Why Intel Chose TSMC Over Its Own Manufacturing Process For Arc Alchemist GPUs
Intel‘s upcoming lineup of Arc Alchemist graphic cards represent the company’s very first serious foray into the discrete GPU market. But, even with such a pivotal product that doesn’t have the room to fail, Intel chose to trust one of its competitors. Instead of manufacturing the GPUs in its own in-house fabs, Intel outsourced the job to semiconductor giant Taiwanese Semiconductor Manufacturing Corporation, better known as TSMC.
While Intel acts as its own chip foundry in most cases—owing up to its vertical strategy, this time the company decided to use TSMC‘s N6 (6nm) node for the job. We already did have a slight understanding of why this was the case, but today we got solid confirmation. ASCII.jp sat down with Intel to learn more about the upcoming Alchemist graphics cards and Intel’s wider future. There, executive Raja Koduri explained as to why Intel chose to go with TSMC for Alchemist.
It is necessary to first determine the process that can be assumed at the start of design. Other features, such as how much operating frequency can be used are also important factors. Cost is also an issue. These three, that is, the cost-performance-capacity, is taken into consideration when deciding which process to use.”
From that quote, we can understand that Intel found the best for Alchemist in TSMC’s N6 node. The N6 node is a leading-edge process that even rivals such as AMD and Nvidia don’t have their hands on yet. That covers performance. Cost and capacity are sort of tied together as Intel would most likely be willing to pay big money if the yields are good enough which, in TSMC’s case, certainly would be.
Intel leveraged several other factors and came to the conclusion that TSMC’s N6 node strikes the perfect balance for what the company has envisioned. Intel simply wasn’t confident enough in manufacturing their own GPUs and knew that competition was extremely tough, so they played the safe bet and went with a foundry that is already well-equipped to handle this on all fronts.
Historically, Intel has always trailed behind the competition when it comes to transistor gate length, even if the dies are denser than the rest. Arc Alchemist needed to hit the homerun on its first try so going with a foundry that has ample experience in that field seemed like the best choice. Intel just didn’t want to play a gamble and risk it by manufacturing the GPU in-house as problems regarding yield could grab them by the neck.
If Intel were to manufacture the GPUs in-house, it would be a daring first for the company. And with a product that basically cannot fail, this was a gamble that Intel did not want to play. Moreover, we know that Intel has struggled with developing a flagship process node of its own for quite some time. Apart from numerous failed attempts to transition from one process node to another smaller one, reports have also shed light on Intel’s 10nm fabrication process producing insufficient yields.
By outsourcing manufacturing to TSMC, Intel essentially rids itself off any pressure. This allows the company to dedicate resources to what it does produce in-house and that’s CPUs. This way, Arc Alchemist gets the best treatment possible and Intel can focus on developing other products whilst not being burdened with capacity concerns for its 10nm process. A win-win situation for Intel at the end.
However, that doesn’t mean that Intel will neccesarily keep struggling with that problem. While Intel decided to go with TSMC this time, there’s no saying if the same will happen for the next generation of Arc. Battlemage, set to succeed Alchemist, is under wraps right now as it’s not releasing before 2023. Therefore, whether Intel will switch over to in-house process nodes for Battlemage or stick with newfound partner TSMC remains yet to be seen.